The new coronavirus pneumonia (Corona Virus Disease 2019, COVID-19) has caused huge losses worldwide. We urgently need effective treatment drugs for COVID-19, but what is the fastest way to discover the drugs?

A way to quickly find antiviral drugs is to find some effective ones against other viruses (such as hepatitis C virus or Ebola virus). It also has antiviral activity against COVID-19. Another method is that we can reasonably and specifically target the protein of severe acute respiratory syndrome coronavirus 2 (SARS-CoV-2), thereby interrupting its life cycle.

The SARS-CoV-2 genome encodes about 25 proteins required by the virus to infect humans and replicate, including the infamous spike protein (S protein) (S protein is recognized at the initial stage of infection Human angiotensin converting enzyme 2), two proteases (which can cleave viruses and human proteins), RNA polymerase (which can synthesize viral RNA), and endoribonuclease (which can cleave RNA). Finding drugs that bind to viral proteins and prevent them from functioning is a logical research direction and the preferred direction of many laboratories.

Although all proteins of SARS-CoV-2 are potential drug targets, some may be easier to find targeted drugs, in part because they play a major role in the virus life cycle and there are no human protein homologs. For example, spike glycoprotein, papain-like protease, chymotrypsin-like main protease, RNA-dependent RNA polymerase. The list of World Protein Data Bank logos for the structure shown is in the supplementary appendix, which can be obtained at NEJM.org. ACE2 means angiotensin converting enzyme 2, NSP means non-structural protein, ORF means open reading frame, RdRP means RNA dependent RNA polymerase.

One way to achieve this goal is to use computational structure-based drug discovery to simulate. In this process, the computer "docks" the test compound into the binding site in the three-dimensional model of the protein target. The binding affinity of the compound is calculated by a physics-based equation that quantifies the interaction between the drug and its target. The top-ranked compounds are then tested experimentally to determine whether they actually bind and produce the desired downstream effects (such as preventing viral infections) on cell and animal models.

The library for docking and changing uses can find compounds that can be used quickly. Larger libraries can contain billions of compounds and are very useful for quickly discovering new compounds that have not been tested in humans.

Structure-based drug discovery methods have played an important role in the discovery of antiviral drugs, such as nelfinavir (nelfinavir) used in the treatment of human immunodeficiency virus (HIV) infections discovered in the 1990s. However, unfortunately, the efficiency of this process was relatively low at that time: the calculation was inaccurate and the computer performance was poor, so only about 100 compounds could be docked at a time. In addition, the target and the drug must be strictly docked with a lock and key. In real life, strict docking is not very common, because the internal heat-driven movement of the protein causes the shape of the binding site to change.

Since the 1990s, the performance of supercomputers has increased by about 1 million times. Now, the strict docking of more than one billion compounds can be completed within a few days. Therefore, the performance of virtual high-throughput screening is superior to the equivalent experimental high-throughput screening, and it can quickly identify very tightly bound compounds. In addition, molecular dynamics simulations can be used to calculate internal protein movements, and in a process called "ensemble docking", the different shapes formed by the binding sites can be used to screen drug candidates. This method is more practical than strict docking and has been successful in HIV drug discovery after 2000.

Modern supercomputers (such as Summit, the Oak Ridge National Laboratory in the United States, the most powerful supercomputer in the world today) can perform massively parallel processing, that is, perform multiple calculations simultaneously. In this way, it is possible to run molecular dynamics simulations of many copies of the target in parallel, and each simulation explores a slightly different conformational space. Therefore, Summit can be used to obtain a comprehensive simulation model of SARS-CoV-2 protein drug targets in a day, and it takes several months to use a general computer cluster. Supercomputers are also used for fast parallel docking of large compound databases. Therefore, the field of structure-based drug discovery is ready to obtain results quickly.

So, what is the situation now? The traditional method of discovering and approving new drugs requires ten years of hard work and is not suitable for the current COVID-19 pandemic. Because the safety of existing drugs is known, changing their use may be a rapid mechanism for using drugs. Therefore, the preprinted paper website published a preliminary report of a collection docking study in mid-February, which used a supercomputer to connect the repurposed compound database to the viral S protein. The study based on the calculated binding to the S protein receptor. The binding affinity between domains ranks 8,000 compounds. The antiviral activity of the top ranked compounds in the initial S protein virtual screening against live viruses is currently being tested. The results will provide a basis for future calculations in a fast and iterative process.

However, in the surreal world where COVID-19 research is accelerating, progress will soon become outdated. Many new experimental three-dimensional structures of S protein and other viral targets are being reported one after another. This process needs to refine and repeat the simulation and docking. Artificial intelligence is being used to predict drug combinations. Different types of laboratory screening programs have been established around the world and are increasing. At the same time, for several SARS-CoV-2 proteins, virtual high-throughput screening and assembly docking pipelines have entered full production mode, which includes screening using supercomputers and using large amounts of cloud computing resources. None of this can guarantee success in a given time frame, but combining reason, scientific insight, and originality with the most powerful tools will provide us with the best chance of success.